New course with Meta: Prompt Engineering with Llama 2

I'm excited to introduce this short course Prompt Engineering with llama, built in partnership with Meta and taught by Ahmed Sanghani, Director of Partner Engineering of the LLAMA team. In this course you'll learn best practices for prompt engineering with Meta's Llama models, which is a collection of powerful LLMs for building generative AI applications. Meta's open source work has been a game changer in generative AI. The fact that developers can download the ways of the LLAMA model has opened up a lot of new use cases. In this course you learn directly from Meta's Amit Sanghani to guide the LLAMA model to the best possible output for your task.

Thank you Andrew. I'm really excited to show you all the exciting things that you can do with the LLAMA models.

In this course you use llama2 for personal everyday tasks that an intelligent virtual assistant might might help you with. These include writing a birthday card for a friend, or brainstorming fun activities to do on the weekend, or extracting the sentiment of text messages as well as summarizing emails, figuring out if you and your friends have enough cars to drive everyone to a restaurant, writing debugging and learning codes, and helping you check if a restaurant review they're writing is safe and inoffensive. In this process you apply prompting techniques such as in context learning, where the model can infer the task you're asking it to complete by you giving it some examples. And you also apply chain of thought prompting to help the LLAMA model think through complex thoughts step by step leading to better and more accurate results. A defining feature of the LLAMA models is that they're available for developers and users like you to freely download, examine, modify, train and use in your own applications. This is really exciting and is leading to broad adoption of these models by startups and large enterprises. Amit has led developer facing engineering teams to help many use LLAMA as well as Pytorch, and the mission of his team is to democratize AI and increase adoption of these platforms by making it easier for developers to integrate them into their products.

Thanks, Andrew. Working with Llama models will give you a lot of flexibility. The course will show you how to run these models on hosted API services, which allows you to easily try out different model versions and determine the best one for your needs. You have more control over the privacy of your data with LLAMA models since your prompts, responses or any data from LLAMA is never shared with Meta. And because developers all over the world are working with the same models, there's a thriving community of users to ask for help and share your questions and ideas with. In addition to the examples that Andrew mentioned, you will also get a chance to try out codelama models which have been trained to understand and write computer code well. If you have never coded before, don't worry about it. Codelama can help. You can ask codelama to help you write a program based on an idea you have, and the model will write all the code for you and even explain how it works. It's a great tool that can help everyone learn to code. You'll learn to use Llama Guard, a special Llama model that you can use to ensure your LLM powered applications run safely. Llama Guard has been specifically trained to detect toxic and harmful language in both user prompts and LLM outputs. I think it's going to be an important tool for building safe, responsibility, responsible AI, and you'll get to try it for yourself in this course.

So this sounds like it'll be a great overview of the Llama models. I routinely run Llama on my personal laptop, so thank you Meta for making these models available and I'm really looking forward to your sharing best practices on using llama. So I hope you take this course and enjoy playing around with the model because I'm sure you find this course to be very illuminating.